Article Network Analysis Identifies Mitochondrial Regulation of Epidermal Differentiation by MPZL3 and FDXR Graphical Abstract Authors Aparna Bhaduri, Alexander Ungewickell, Lisa D. Boxer, Vanessa Lopez-Pajares, Brian J. Zarnegar, Paul A. Khavari Correspondence [email protected] In Brief To identify non-transcription factor regulators of epidermal differentiation, Bhaduri et al. devised a network approach called proximity analysis and applied it to identify MPZL3 as a mitochondrial protein required for epidermal differentiation. MPZL3 promotes the enzymatic activity of mitochondrial protein FDXR, and the factors together regulate ROS-mediated epidermal differentiation. Highlights d Proximity analysis successfully reconstructs expression- based networks in eukaryotes d Proximity analysis identifies MPZL3 as highly connected in epidermal differentiation d MPZL3 is required for differentiation and localizes to the mitochondria d MPZL3 binds FDXR, and together they regulate ROS required for differentiation Bhaduri et al., 2015, Developmental Cell 35, 444–457 November 23, 2015 ª2015 Elsevier Inc. http://dx.doi.org/10.1016/j.devcel.2015.10.023 Developmental Cell Article Network Analysis Identifies Mitochondrial Regulation of Epidermal Differentiation by MPZL3 and FDXR Aparna Bhaduri,1 Alexander Ungewickell,1,2 Lisa D. Boxer,1,3 Vanessa Lopez-Pajares,1 Brian J. Zarnegar,1 and Paul A. Khavari1,4,* 1Program in Epithelial Biology, Stanford University School of Medicine, Stanford, CA 94305, USA 2Division of Hematology, Stanford University, Stanford, CA 94305, USA 3Department of Biology, Stanford University, Stanford, CA 94305, USA 4Veterans Affairs Palo Alto Healthcare System, Palo Alto, CA 94304, USA *Correspondence: [email protected] http://dx.doi.org/10.1016/j.devcel.2015.10.023 SUMMARY differentiation, may be also regulated beyond the nucleus, including in the mitochondria, the lysosome, and the ribosome Current gene expression network approaches (Kasahara and Scorrano, 2014; Settembre et al., 2013; Xu commonly focus on transcription factors (TFs), et al., 2013; Xue and Barna, 2012). Understanding how these biasing network-based discovery efforts away from levels of regulation interact with canonical nuclear gene regula- potentially important non-TF proteins. We developed tion processes can better illuminate how differentiation pro- proximity analysis, a network reconstruction method ceeds through feedback and feedforward loops and may that uses topological constraints of scale-free, provide additional insight into treatments for congenital and somatic diseases of deranged differentiation. small-world biological networks to reconstruct rela- Gene expression network reconstruction is a commonly used tionships in eukaryotic systems, independent of sub- method of describing organism level systems (Herrga˚ rd et al., cellular localization. Proximity analysis identified 2008; Tong et al., 2004), identifying key relationships that can MPZL3 as a highly connected hub that is strongly highlight regulators of a process (Carro et al., 2010), and delin- induced during epidermal differentiation. MPZL3 eating cell types at a transcriptional level (Yang et al., 2013). was essential for normal differentiation, acting down- While the applications of network-based analyses are vast, cur- stream of p63, ZNF750, KLF4, and RCOR1, each of rent network algorithms favor models of less complex organisms which bound near the MPZL3 gene and controlled such as Escherichia coli or networks with a bias toward known its expression. MPZL3 protein localized to mitochon- transcription factors. The recent DREAM5 consortium analysis dria, where it interacted with FDXR, which was of the highest performing, most used network reconstruction itself also found to be essential for differentiation. algorithms highlighted that while combining the outputs from multiple existing network algorithms improved upon the perfor- Together, MPZL3 and FDXR increased reactive mance of a single algorithm alone, the ability to reconstruct oxygen species (ROS) to drive epidermal differentia- known relationships fell significantly from in silico networks to tion. ROS-induced differentiation is dependent upon E. coli networks to eukaryotic Saccharomyces cerevisiae net- promotion of FDXR enzymatic activity by MPZL3. works (Marbach et al., 2012). Additional deconvolution efforts ROS induction by the MPZL3 and FDXR mitochon- aimed to improve these metrics, but were only able to do so in drial proteins is therefore essential for epidermal an incremental manner for eukaryotes (Feizi et al., 2013). There- differentiation. fore, significant challenges persist in using network reconstruc- tion approaches to understand human tissue differentiation, especially when searching beyond transcription factors. INTRODUCTION The epidermis is an excellent model for the application of a network reconstruction approach to discover non-transcription Somatic differentiation requires the concurrent regulation of a factor regulators because it is a relatively well characterized tis- large number of genes that are necessary for the tissue-specific sue with in vitro model systems derived from primary human functions of differentiated cells. Much of this regulation comes cells. The epidermis is comprised of a basal layer of progenitor from transcription factors that regulate gene expression, cells that give rise to the layers of epidermis that exit the cell silencing genes that maintain a proliferative cell-cycle state cycle, enucleate, and provide barrier function through expres- and activating genes that are necessary for the cellular functions sion of tissue-specific differentiation genes. The transcriptional characteristic of the differentiated tissue. However, only a frac- master regulator of the epidermis, p63 (Mills et al., 1999; Truong tion of the genes that are induced during a somatic differentiation et al., 2006; Yang et al., 1999), targets key genes including program are nuclear factors, and recent work suggests that ZNF750 (Boxer et al., 2014; Sen et al., 2012) and MAF/MAFB (Lo- differentiation processes, as well as disease-causing defects in pez-Pajares et al., 2015). Other transcription factors implicated 444 Developmental Cell 35, 444–457, November 23, 2015 ª2015 Elsevier Inc. in the regulation of epidermal differentiation include KLF4 (Patel straints is that by simulating the preferential attachment models et al., 2006; Segre et al., 1999), GRHL3 (Hopkin et al., 2012; Yu that have been described to result in scale-free networks (Bara- et al., 2006), and OVOL1 (Teng et al., 2007). Recent work gener- basi and Albert, 1999), the relationships between genes that are ated kinetic gene expression data in the regeneration of differen- biologically related will be highlighted by higher probabilities tiated epidermal tissue (Lopez-Pajares et al., 2015). The ability to within this network. We first optimized proximity analysis for use this type of kinetic data in a model with well-characterized various input parameters on a series of in silico networks gener- positive controls makes it an ideal system for applying network ated by DreamNetWeaver (Schaffter et al., 2011)(Figure S1C). reconstruction approaches to discover new regulators. We used proximity analysis to analyze the Saccharomyces cer- Here, we develop and apply proximity analysis to network evisiae expression data provided to the DREAM5 network chal- reconstruction during the process of epidermal differentiation. lenge (Table S1) and compared our output to the top-performing Analyzing a time course of gene expression during differentiation algorithms that have generated networks from this data (Feizi generated a network of strongly connected genes, including et al., 2013; Marbach et al., 2012). Because of our interest in those with known roles in differentiation as well as novel candi- identifying non-transcription factor regulators, we used an alter- date regulators. A top hit, MPZL3, is highly induced in the native evaluation metric to the binding site derived metrics used process of epidermal differentiation and downregulated in in the DREAM5 challenge. Using validated gene ontology (GO) cutaneous squamous cell carcinoma. MPZL3 was found to classifications, a frequently used conservative network valida- be essential for epidermal differentiation. Its expression was tion metric (Chagoyen and Pazos, 2010), proximity analysis out- controlled by several known transcriptional regulators of differ- performed the other examined algorithms in both the number entiation, including p63, ZNF750, KLF4, and RCOR1. Live-cell and fraction of edges belonging to the same GO term (Fig- vicinal protein labeling followed by mass spectrometry demon- ure S1D). We additionally noted that while all algorithms exam- strated that MPZL3 primarily interacts with mitochondrial pro- ined had an average local clustering coefficient characteristic teins, with mitochondrial localization confirmed by electron of a small-world network and a power law degree distribution microscopy. Among MPZL3-interacting proteins was FDXR, a characteristic of a scale-free network topology (Figure S1E), mitochondrial enzyme that catalyzes the reduction of ferredoxin. proximity analysis was the most small-world (Figure S1F) with We observed that FDXR is also required for normal epidermal dif- the smallest residual when compared to a perfect power law dis- ferentiation, and its ectopic expression is capable of rescuing the tribution (Figure S1G). These data suggest that in a eukaryotic differentiation